JumpCloud survey finds lower AI confidence as agents reach production

JumpCloud said self-reported AI maturity among IT leaders fell to 23% from 40% over six months, based on a survey of 800 IT leaders in the U.S. and U.K. The company argues the drop reflects a shift from pilots to production, where AI agents start touching real systems and expose gaps in governance, identity and oversight.

No funding, valuation or revenue figures are involved. The numbers come from JumpCloud’s Q3 2026 AI Readiness Research report, a vendor-produced study presented alongside commentary from Rajat Bhargava, the company’s CEO and co-founder. The findings are self-reported, so they measure how IT leaders rate their own organizations rather than an independent audit of AI operations.

Production is changing the risk profile

According to JumpCloud, the decline in confidence is concentrated among organizations that have moved AI agents beyond test environments. The company said those teams are not pulling back from AI, but are finding that production use requires controls that pilots did not demand.

JumpCloud reported that 84% of organizations plan to increase AI use in IT operations over the next six to 24 months. That suggests the lower maturity rating is not a broad retreat from AI adoption. It is a reassessment of how ready companies are to run autonomous or semi-autonomous systems inside existing workflows.

The distinction matters for operators. A pilot can be limited to a narrow task with a defined data set and close supervision. In production, an agent may access operational systems, make workflow decisions and run continuously. JumpCloud said that raises questions many teams did not have to answer during pilots, including whether they can see every agent in use, know what each can access and identify unexpected behavior after it occurs.

Identity controls are the weak point

The most concrete finding in the report concerns non-human identities. JumpCloud said non-human identity governance was the least adopted AI security practice it measured, with adoption at 21% of organizations.

The company also reported that non-human identities outnumber human users in 83% of organizations. JumpCloud defined the problem as one of accountability: agents and service-like accounts may continue operating after their original purpose has expired, retain permissions and lack a named owner or offboarding process.

JumpCloud calls these “Zombie Agents.” The term is vendor branding, but the operational issue is familiar to security and IT teams. AI agents add another class of machine identity to environments where permissions and ownership are already hard to track.

Consolidation and measurement separate mature teams

JumpCloud said organizations in the top tier of its maturity model are five times more likely than the average organization to report no barriers to expanding AI agents. The company attributed that difference to several practices: consolidating IT environments, treating AI agents as governed identities and measuring outcomes rather than counting deployments.

The report does not provide a public breakdown in the cited summary of how many respondents fall into each maturity tier. It also does not establish that those practices alone caused better results. The findings do, however, point to a practical issue for companies selling and buying AI systems: deployment metrics are becoming less useful without evidence that access, ownership and accountability are in place.

For vendors, the survey supports a shift in enterprise AI spending toward governance and infrastructure around agents, not just the agents themselves. For CIOs and security leaders, JumpCloud’s data frames lower confidence as a sign that teams are finding the operational costs of production AI, rather than assuming early pilots proved readiness.

This story draws on original reporting from VentureBeat.